Toward Fast and Accurate Vehicle Detection in Aerial Images Using Coupled Region-Based Convolutional Neural Networks

摘要: Vehicle detection in aerial images, being an interesting but challenging problem, plays important role for a wide range of applications. Traditional methods are based on sliding-window search and handcrafted or shallow-learning-based features with heavy computational costs limited representation power. Recently, deep learning algorithms, especially region-based convolutional neural networks (R-CNNs), have achieved state-of-the-art performance computer vision. However, several challenges limit the applications R-CNNs vehicle from images: 1) vehicles large-scale images relatively small size, poor localization objects; 2) particularly designed detecting bounding box targets without extracting attributes; 3) manual annotation is generally expensive available training not sufficient number. To address these problems, this paper proposes fast accurate framework. On one hand, to accurately extract vehicle-like targets, we developed accurate-vehicle-proposal-network (AVPN) hyper feature map which combines hierarchical maps that more object detection. other propose coupled R-CNN method, AVPN attribute network vehicle's location attributes simultaneously. For original annotations, use cropped image blocks data augmentation avoid overfitting. Comprehensive evaluations public Munich dataset collected demonstrate accuracy effectiveness proposed method.